A Data-Driven Method for Assessing Carbon Emissions from Cement Production

Abstract

This invention belongs to the field of environmental assessment technology for production, specifically disclosing a digitally driven carbon emission assessment method for cement production. This method first collects real-time operational characteristics and carbon emission intensity data for each production unit, and then uses cluster analysis to divide the production into groups of similar processes. Next, it statistically analyzes the carbon emission intensity distribution of each group to dynamically generate performance benchmark values. Then, it calculates the relative deviation between the real-time carbon emission intensity of a specified unit and its corresponding benchmark, and analyzes the historical trend of this deviation to obtain the trend slope. Finally, it combines the relative deviation and the trend slope to calculate a carbon emission exceedance trend index. When this index exceeds a preset threshold, an early warning is automatically triggered and a diagnostic report is generated. This invention achieves a multi-process, integrated carbon efficiency assessment, overcoming the limitations of traditional single-monitoring-point post-event accounting, while also effectively improving the accuracy and foresight of carbon emission management.

Description

Technical Field

[0001] This invention belongs to the field of environmental assessment technology for production, and specifically relates to a digitally driven method for assessing carbon emissions from cement production. Background Technology

[0002] The cement production process is highly complex, with carbon emissions occurring throughout the entire process, from raw material crushing to cement grinding. To reduce environmental impact and align with green production principles, it is necessary to assess the carbon emissions from this production process.

[0003] Existing technologies, such as the online monitoring process system and uncertainty quantification method for carbon emissions from cement production disclosed in Chinese invention patent application number 202211337529.3, involve standardizing the selection of measurement points, assessing the complexity of the flow field in stages, collecting data on flow velocity, CO2 concentration, temperature, pressure, and humidity, and calculating the independent uncertainties of each parameter. A comprehensive model is then established to quantify and synthesize these uncertainties, ultimately achieving a scientific evaluation and assurance of the reliability of online carbon emission data.

[0004] The aforementioned existing technologies mainly focus on the selection of measurement points and the calculation of the uncertainty of each parameter, which belong to the hardware and accuracy verification level. For carbon emission assessment, there are still the following problems: 1. It only verifies the data accuracy of a single monitoring point and fails to link carbon emissions with the production and operation status of multiple processes and the whole process of the enterprise, and cannot provide cross-process carbon efficiency assessment and optimization decision support from the management level.

[0005] 2. The output results are only historical instantaneous data after accuracy calibration, which is essentially a post-event accounting. Without trend prediction and combining the trend prediction results, it is impossible to achieve early warning and forward-looking regulation. Summary of the Invention

[0006] In view of this, in order to solve the above problems, a digitally driven method for assessing carbon emissions from cement production is proposed.

[0007] The objective of this invention can be achieved through the following technical solution: This invention provides a method for assessing carbon emissions from cement production based on digital intelligence. The method includes: real-time collection of operational characteristic data and carbon emission intensity of multiple cement production units; and, based on the operational characteristic data, dividing the multiple cement production units into several peer groups through cluster analysis.

[0008] The historical carbon emission intensity data distribution of each cement production unit within each peer group is statistically analyzed, and the performance benchmark value of that peer group is dynamically generated accordingly.

[0009] Calculate the relative deviation between the real-time carbon emission intensity of a specified cement production unit and the corresponding performance benchmark value of its peer group.

[0010] The historical relative deviation data of the specified cement production unit over multiple consecutive periods are obtained, and the trend slope is obtained by linear regression analysis.

[0011] By combining the current relative deviation with the trend slope, a carbon emission exceedance trend index is calculated.

[0012] When the carbon emission exceedance trend index exceeds a preset risk threshold, an early warning signal is triggered and a management diagnostic report is generated.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) The present invention collects cement production and operation characteristic data in real time and divides the production unit into peer groups with similar process equipment based on cluster analysis, thereby realizing cross-process linkage evaluation of carbon emission intensity under comparable benchmark, thus overcoming the limitations of traditional single monitoring point data, and providing enterprises with a comprehensive comparison and optimization decision basis for carbon efficiency covering multiple production links.

[0014] (2) By introducing the determination of stable operating conditions and performing differentiated preprocessing of different data, this invention effectively filters out the interference of production fluctuations and abnormal events on data quality, ensures the robustness of clustering and benchmark calculation, and also improves the accuracy and operability of carbon efficiency assessment results.

[0015] (3) This invention achieves real-time benchmarking and gap quantification of carbon emission performance by dynamically generating performance benchmark values ​​of peer groups and calculating real-time relative deviation. It overcomes the problem of assessment lag caused by changes in production plans or seasonal factors in traditional static benchmarks, enabling enterprises to quickly identify the actual position of their current carbon emission level in similar production units and providing real-time data support for accurately locating carbon efficiency shortcomings.

[0016] (4) The present invention adopts an adaptive selection of benchmark value generation strategy based on skewness coefficient, which can automatically adjust the baseline reference line according to real-time operation data, effectively adapt to the normal fluctuation of carbon emission intensity caused by changes in operating conditions such as production load and raw material ratio, avoid the evaluation deviation caused by using fixed benchmark value, and ensure the timeliness and fairness of carbon efficiency benchmarking results.

[0017] (5) This invention combines historical deviation to perform trend slope analysis and constructs a carbon emission exceedance trend index, transforming static ex-post accounting into dynamic risk warning, realizing the forward-looking identification and regulation of carbon emission exceedance risk, and also realizing the transformation from single-point static assessment to continuous dynamic tracking, so that the assessment results of carbon emission performance not only reflect the current gap, but also reveal its direction and speed of change. Attached Figure Description

[0018] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a schematic diagram of the overall implementation process of the present invention.

[0020] Figure 2 This is a simplified flowchart of the overall implementation process for the preprocessing of operational feature data in this invention.

[0021] Figure 3 This is a schematic diagram of the quality assessment index calculation process of the present invention.

[0022] Figure 4 This is a schematic diagram of the dynamic performance benchmark value generation process of the present invention.

[0023] Figure 5 This is a schematic diagram of the carbon emission exceedance trend index calculation process of the present invention. Detailed Implementation

[0024] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0025] This invention provides a digitally driven method for assessing carbon emissions from cement production. Please refer to [link / reference]. Figure 1 As shown, the method specifically includes steps S1 to S6, which specifically demonstrates the overall process of the present invention, including data acquisition, cluster analysis, benchmark generation, deviation calculation, trend analysis, and early warning.

[0026] S1. Real-time collection of operational characteristic data and carbon emission intensity of multiple cement production units. Based on the operational characteristic data, the multiple cement production units are divided into several peer groups through cluster analysis.

[0027] Specifically, the operational characteristic data includes production process parameters, material consumption data, output data, equipment uptime data, and current production schedule.

[0028] The equipment operating time data includes the equipment start and stop times, the longest continuous operating time, the frequency of unplanned shutdowns, the duration of unplanned shutdowns, and the production stage in which the unplanned shutdowns occurred.

[0029] The production process parameters, material consumption data, output data, and equipment operating time can be extracted from the cement company's DCS, MES, and energy management system.

[0030] In one specific embodiment, the production process parameters include kiln speed, kiln head negative pressure, kiln tail temperature, decomposer outlet temperature, preheater outlet temperature and pressure at each stage, air pressure in each chamber of the grate cooler and clinker temperature, main motor current of the raw meal mill or cement mill, classifier speed and inlet / outlet pressure difference, etc. Material consumption data include instantaneous and cumulative values ​​of raw meal input, coal input, and blending material addition. Production data include hourly clinker output and hourly cement mill output, etc.

[0031] Further, please refer to Figure 2 The process includes preprocessing the operational characteristic data before cluster analysis, and the preprocessing process includes steps A1-A7.

[0032] A1. Monitor the variance of coal feed rate fluctuation and kiln drive load fluctuation of the production unit, and calculate the quantile values ​​of the two in their respective historical variance datasets.

[0033] A2. If the duration of the quantile values ​​being continuously lower than the first threshold of their respective historical fluctuation variances exceeds one complete production shift, the current time period is determined to be a stable operating condition interval.

[0034] Understandably, the first threshold can be the 30th quantile value from the historical variance dataset. This step is used to eliminate the dimensional effects caused by fluctuations in raw material composition and adjustments to operating conditions, ensuring that the data reflects the deviation from a stable baseline.

[0035] A3. For the production process parameters, material consumption data, and output data, use the mean and standard deviation calculated from historical data samples within the stable operating range to perform Z-score standardization and output standardized data.

[0036] A4. Input the equipment start-up and shutdown times and the current production schedule into the parser. The parser classifies the times into the corresponding production modes by matching event tags and timestamps, and dynamically assigns a cycle mapping function to each mode.

[0037] It should be noted that using sine and cosine transformation on periodic features such as equipment start-up and shutdown times can map linear time values ​​to cyclic trigonometric function values, allowing machine learning algorithms to understand the periodic patterns of time, such as identifying the differences between night shifts and day shifts. This avoids incorrectly classifying cross-day time points as distant or irrelevant data points, ensuring the accuracy of clustering based on the similarity of work periods.

[0038] For example, taking 23:59 and 00:01 as cross-day times, directly representing the time points with numbers from 0 to 24 and normalizing them would prevent the algorithm from understanding that 23:59 and 00:01 differ by only 2 minutes in the real world. Instead, it would consider 23.99 and 0.01 to be two unrelated points with vastly different values. In contrast, after conversion, the sine and cosine values ​​corresponding to 23:59 and 00:01 are almost equal. Subsequent clustering algorithms can then correctly cluster production units that always operate during the night shift into one group and those that always operate during the day shift into another.

[0039] It should be added that the parser uses natural language processing (NLP) technology to analyze the event descriptions in the production schedule. By matching keywords and semantic rules, it classifies events into three modes: normal production, planned maintenance, and low-load operation. NLP technology is a relatively mature technology and will not be elaborated upon here.

[0040] Preferably, dynamically assigning a corresponding periodic mapping function to each type of mode includes: assigning a standard 24-hour periodic sine and cosine mapping function to the normal production mode.

[0041] Assign a feature generation function to the planned maintenance mode: take the maintenance start time as the time origin, and calculate the elapsed time length of the equipment start and stop times relative to the origin as the feature value.

[0042] Assign a 48-hour period sine-cosine mapping function to the low-load operation mode.

[0043] A5. Based on the production mode to which the equipment starts and stops, call the corresponding cycle mapping function to calculate its transformed characteristic value.

[0044] A6. Using the unplanned downtime duration and the current production stage as input, calculate the downtime interference weight coefficient through downtime impact quantification. Based on the weight coefficient, calculate the unplanned downtime frequency and duration through linear weighting and perform minimum-maximum normalization.

[0045] Understandably, using min-max normalization for numerical features such as unplanned downtime frequency and duration can make features with different units of downtime frequency and duration comparable in importance and contribution in cluster analysis, thereby achieving a fair and effective multi-dimensional assessment of equipment stability.

[0046] Among them, the quantitative calculation of the downtime impact can be obtained online through the downtime impact quantitative model. This model is trained with historical data and takes the production stage, duration, and increase in unit product energy consumption within a certain period after the downtime event as inputs. It outputs a weight value that characterizes the actual impact of the downtime on the overall energy efficiency.

[0047] In one specific embodiment, the formula for the downtime impact quantification model is as follows: , This indicates the overall impact weight of the shutdown event. The larger the value, the greater the negative impact of the shutdown event on carbon emissions, making shutdown events with a greater impact on carbon emissions occupy a more important position in data analysis.

[0048] in, This is a basis function trained on historical data, reflecting the impact of downtime per minute on average carbon emission intensity at different production stages. Its function value represents the base effect of downtime per unit time on global carbon emission intensity at that production stage. It is a weighted amplifier, specifically trained using machine learning regression models such as linear regression and gradient boosting trees. Using the production stage, duration, and subsequent carbon emission intensity changes of historical downtime events as training data, it fits a quantitative relationship between downtime per unit time at different production stages M and subsequent changes in carbon emission intensity per unit of product. The output is a real number greater than 0. The higher the value, the more severe the impact of downtime per unit time on carbon emissions during that production phase.

[0049] This represents the duration-based impact function, used to quantify the amplification effect of downtime on the overall impact. The effect of downtime duration increases with time, with a significant initial impact followed by a gradual slowdown in the rate of increase. This refers to unplanned downtime. The time scale constant, with units of and . Consistency, used to normalize time, for example... It can be set to 60 minutes or 120 minutes. It is a logarithmic function with base 10, when In a very short time, Slightly greater than 1, its logarithm is also small, and it grows relatively quickly. While the logarithmic function grows slowly over long periods, it effectively avoids the problem of excessive weighting of a single extremely long shutdown event, which could suppress other features. Adding 1 is a smoothing term, ensuring that when... As the value approaches zero, the function value approaches zero rather than negative infinity, ensuring mathematical rationality. This reasonably represents the impact of downtime, acknowledging that the impact increases with time while avoiding the weight distortion problem that may arise from linear growth.

[0050] A7. Integrate standardized data, feature values, and normalization results to form a preprocessed dataset.

[0051] Understandably, cement production is a continuous and complex process. Frequent equipment start-ups and shutdowns, unplanned downtime, and changes in production schedules can introduce significant noise into the data. Directly using raw production and operation data for cluster analysis can lead to distorted results due to inherent fluctuations and anomalies in the production process.

[0052] It should be noted that identifying stable operating condition ranges ensures the consistency of data benchmarks used for subsequent comparisons. Setting up a production scheduling parser and cycle mapping function normalizes equipment status data at different time points to comparable dimensions. Quantifying the downtime interference weight eliminates the interference of non-productive factors on energy efficiency assessment. Ultimately, this combined preprocessing method outputs high-quality feature data that has been cleaned, aligned, and weighted, laying a solid foundation for constructing a scientifically sound peer group. This ensures the objectivity of dynamic performance benchmarks and the effectiveness of early warning signals for carbon emission exceedance trends.

[0053] Furthermore, the cluster analysis process includes steps B1-B5.

[0054] B1. Use the preprocessed operational feature data as a feature vector, assign differentiated weight coefficients according to the production process to which the feature dimension belongs, and construct a weighted feature vector.

[0055] Preferably, the weighting coefficient allocation rule is as follows: assign weights to feature dimensions belonging to the clinker calcination process. .

[0056] Weights are assigned to feature dimensions belonging to the raw material preparation and cement grinding processes. .

[0057] Assign weights to feature dimensions that belong to auxiliary processes. ,and , .

[0058] In one specific embodiment, clinker calcination is the core energy-consuming and carbon-emitting stage in cement production, accounting for approximately 85%-95% of the total carbon emissions from fuel combustion and carbonate decomposition. Its weight can be set to 0.6. Raw material preparation and cement grinding primarily consume electricity, representing indirect carbon emissions. While their carbon emissions are much lower than clinker calcination, they remain significant energy-consuming stages in production. A moderate weight is assigned to these stages to differentiate units with different grinding efficiencies; a weight of 0.3 is suitable. Auxiliary processes such as in-plant transportation, lighting, and ventilation account for a very small proportion of energy consumption and carbon emissions. A lower weight is assigned to these processes to capture differences under extreme conditions, but to avoid them having a decisive impact on the grouping results; a weight of 0.1 is suitable.

[0059] It should be noted that by assigning the highest weight to the clinker calcination process, the second highest weight to the raw material preparation and cement grinding processes, and the lowest weight to auxiliary processes, units with similar operating models in key energy consumption and emission stages can be prioritized and grouped together. This effectively avoids misclassifying production units with vastly different core processes as competitors due to similar secondary characteristics, greatly improving the scientific rigor and relevance of competitor grouping, and providing a reliable guarantee for establishing meaningful carbon emission performance benchmarks in the future.

[0060] B2. Based on the range of typical process route combinations in the cement industry, a set of candidate cluster quantities is preset. For each candidate quantity, a clustering process is performed based on the weighted feature vector and a quality assessment index is calculated.

[0061] Please see Figure 3 As shown, the specific calculation process of the quality assessment index includes steps B21-B27.

[0062] B21. For each candidate number in the candidate cluster quantity set, select the initial number of cluster centers based on the cement production equipment configuration constraints. The constraints include that the production units corresponding to the initial cluster centers are consistent in kiln type configuration and are at the same level in terms of production capacity.

[0063] For example, the initial cluster center selection rule is as follows: calculate the shortest Euclidean distance between each data point and the selected cluster center, and select the next cluster center with a probability proportional to the square of the Euclidean distance.

[0064] B22. Calculate the weighted Euclidean distance from each weighted feature vector to each cluster center, and use it as a similarity measure for iterative clustering.

[0065] B23. After each iteration, check the compatibility of production processes within each cluster and reallocate production units with process conflicts.

[0066] B24. After the clustering process converges, calculate the average profile coefficient of the clustering result corresponding to the current number of clusters and the variance of the carbon emission intensity among the clusters in the clustering result.

[0067] Understandably, the average silhouette coefficient is used to determine the compactness of samples within a cluster and the distinctness of samples between clusters, with a value ranging from -1 to 1. A coefficient closer to 1 indicates better clustering, while a coefficient closer to 0 indicates high overlap between clusters. A coefficient close to -1 indicates clustering errors. The average silhouette coefficient is calculated by combining the average Euclidean distance within clusters and the minimum average Euclidean distance between clusters, and the formula used is an existing formula, which will not be shown here. The inter-cluster variance is used to quantify the degree of difference in carbon emission intensity between different clusters. A larger variance indicates a more significant difference in carbon emission intensity levels between clusters, while a smaller variance indicates that the carbon emission intensity between clusters is closer, and the clustering is less effective at distinguishing differences in carbon emission intensity.

[0068] B25. Perform a statistical significance test on the variance of carbon emission intensity between clusters in the clustering results, and obtain the test conclusion.

[0069] Understandably, the inter-cluster differences obtained through variance calculations may be due to random fluctuations. For example, a randomly selected sample might happen to result in large differences in cluster means. The purpose of the significance test is to determine whether the differences in carbon emission intensity between clusters are real statistical differences rather than random errors, thereby verifying the validity of the clustering results. Specifically, this can be verified using the F-test, a common existing significance testing technique. The specific testing process and criteria for determining whether the test passes will not be detailed here.

[0070] B26. If the statistical significance test fails, the quality assessment index will be assigned a value of zero.

[0071] B27. If the statistical significance test is passed, the mean profile coefficient and the variance are subjected to minimum-maximum normalization, and the final quality assessment index is output by linear weighted fusion.

[0072] It should be noted that a larger average profile coefficient indicates more similar operating modes among production units within the same cluster, while a larger variance in carbon emission intensity between clusters indicates more significant differences in carbon emission performance levels between different clusters, and the clustering results are better able to distinguish groups with different energy efficiency levels. By combining these two dimensions, the optimal clustering scheme can be selected that reflects the differences in actual operating modes and has significant distinguishability in carbon emission performance.

[0073] It is understood that the ultimate goal of this invention is to conduct effective carbon emission assessment, that is, to assign a higher weight, for example, 0.6 to 0.7, to the variance of carbon emission intensity reflecting business differentiation, while assigning a relatively lower weight, for example, 0.4 to 0.3, to the average profile coefficient reflecting the geometric quality of clustering. The weights can be adjusted according to actual business needs, with priority usually given to carbon emission differentiation; the above ranges are empirical values.

[0074] It's understandable that common clustering evaluation metrics, such as the silhouette coefficient, often only focus on the density of data points in geometric space, but cannot guarantee that the clustering results have practical significance under specific carbon emission assessments. To avoid invalid groupings that cannot effectively distinguish the carbon emission levels of different groups during the clustering process, the silhouette coefficient and the variance of carbon emission intensity between clusters are combined.

[0075] It should be noted that by mandating that the differences in carbon emission intensity between clusters pass a statistical significance test, and using this as a veto factor in quality assessment, it is ensured that each group truly exhibits a fundamental difference in carbon emission performance. Finally, the weighted fusion of the average silhouette coefficient, reflecting the geometric quality of the clusters, and the normalized result of the carbon emission intensity variance, reflecting business differentiation, ensures that the selected optimal number of clusters simultaneously satisfies the dual objectives of high intra-group similarity and significant inter-group differences. This provides the most scientific data grouping basis for establishing dynamic performance benchmarks with practical guiding significance.

[0076] B3. The number of candidate clusters with the highest quality assessment index is taken as the target number of clusters.

[0077] B4. Based on the target number of clusters and the weighted feature vector, perform a clustering process that includes process compatibility verification, and count the mode of kiln type, mill model and capacity scale within each cluster.

[0078] B5. Generate process labels in the format of capacity scale-kiln type-mill type, and then classify the cement production units according to the process labels.

[0079] Understandably, the carbon emission intensity of a cement production unit is fundamentally influenced by its kiln type, mill type, and production capacity. Simply comparing all production units together without differentiation would lead to biased and unfair conclusions. Therefore, it is essential to first scientifically group production units based on these key technological characteristics, ensuring inherent comparability within each group. This step, through the introduction of differentiated weights, process compatibility verification, and automated process label generation, aims to embed industry expertise into data-driven algorithms, thereby overcoming the drawbacks of biased comparisons and laying the foundation for subsequent carbon emission performance benchmarking on a level playing field.

[0080] It should be noted that the process labels generated based on production capacity scale, kiln type, and mill type ensure that members within each peer group have highly similar processes in essence. This greatly improves the accuracy and fairness of the subsequently generated dynamic performance benchmarks, allowing the comparison of carbon emission intensity and the calculation of deviation to truly reflect the level of production operation and management, rather than being dominated by inherent equipment differences, thereby ensuring the validity and credibility of the conclusions of the entire assessment method.

[0081] S2. Statistically analyze the distribution of historical carbon emission intensity data for each cement production unit within each peer group, and dynamically generate the performance benchmark value for that peer group based on this data.

[0082] Specifically, please refer to Figure 4 As shown, the specific process of generating the dynamic performance benchmark value includes steps S21-27.

[0083] S21. For each peer group, collect the carbon emission intensity data sequence of all cement production units within the group within the preset statistical period.

[0084] S22. Calculate the standard deviation of the sequence. If it exceeds the preset stability threshold, then filter the carbon emission intensity data sequences of the top preset percentage of cement production units.

[0085] Understandably, when the standard deviation of carbon emission intensity data series within a peer group exceeds a stability threshold, the system will typically automatically select the top 30% of production units with the lowest carbon emission intensity within that group, using their carbon emission intensity data series as the basis for calculations. Furthermore, the preset stability threshold can be based on industry experience.

[0086] It should be noted that when there are significant differences in carbon emission levels among production units within a peer group, directly using data from the entire group to calculate the benchmark would result in the outcome being averaged across a large number of units at medium or low levels, failing to incentivize advanced units and spur on laggards. Using data from the top 30% of leading units ensures that the benchmark effectively drives management actions aimed at improving carbon efficiency. Simultaneously, the 30% sample size ensures data representativeness, preventing the benchmark value from being easily influenced by a single outlier due to a small sample size, thus enhancing the stability of the benchmark.

[0087] S23. Based on the filtered data, calculate the mean, standard deviation, first specific quantile value and second specific quantile value, where the first specific quantile value is greater than the second specific quantile value, and determine the data distribution pattern.

[0088] In a preferred implementation, the first specific quantile value can be the 25th or 30th quantile. This means that the benchmark value represents a threshold for the top 25% or top 30% of the peer group's advanced level. The second specific quantile value can be the 15th or 10th quantile. That is, when the dispersion is significant, a higher level of quantile value is used.

[0089] It is understood that the first specific quantile value is not limited to the 25th or 30th quantile. Those skilled in the art can set it according to specific policy requirements or industry standards, such as setting it to the 20th quantile to represent a more advanced level, or setting it to the 40th quantile as a phased target. The core is to use the statistical characteristics of quantiles to dynamically set a relatively advanced benchmark. The second specific quantile value is similar and can be adjusted accordingly based on the actual situation.

[0090] It should be added that the specific steps for determining the data distribution pattern include: calculating the skewness coefficient based on the carbon emission intensity of the cement production unit, the mean and standard deviation of the screened data sequence.

[0091] When the skewness coefficient is greater than the preset positive skewness threshold, it is determined to be a strongly positively skewed distribution.

[0092] When the absolute value of the skewness coefficient is less than the preset normal threshold, it is determined to be close to a normal distribution.

[0093] When the skewness coefficient is between the preset normal threshold and the preset positive skewness threshold, it is determined to be a slightly positively skewed distribution.

[0094] The skewness coefficient can be calculated based on the formula for the third central moment, specifically: , Indicates the skewness coefficient. This indicates the cement production unit number. , Indicates the first Carbon emission intensity of a single cement production unit This represents the average value of the filtered data sequence. This represents the standard deviation of the filtered data sequence. This indicates the number of cement production units.

[0095] Understandable, This is an unbiased correction factor for sample size. When the sample size is small, this correction makes the calculated skewness coefficient closer to the true population skewness, resulting in a more accurate estimate. When the sample size is large, this factor is the reciprocal of the sample size, and its influence decreases. Using this factor for small sample correction ensures the stability of calculations under different sample sizes, providing a reliable basis for determining the distribution pattern in the scientific setting of benchmark values.

[0096] The third central moment is the formula for calculating the sum of cubes of the deviation of data from the mean. It is a key part of reflecting the degree of distribution skewness. By calculating the third central moment, the asymmetry of data distribution can be accurately captured. Skewness is a measure of the degree of asymmetry in the distribution of data.

[0097] In one specific embodiment, the positive bias threshold is preferably set to 0.5, and the normal threshold is preferably set to 0.3.

[0098] S24. When the data is strongly positively skewed, the first specific quantile value is used as the initial performance benchmark value. When the data is slightly positively skewed, the initial performance benchmark value is output based on the weighted mixture calculation of the first specific quantile value and the average value. When the data is close to normal, the average value is used as the initial performance benchmark value.

[0099] The weighted mixing calculation formula is as follows: , As the initial performance benchmark value, To preset the positive bias threshold, This represents the difference between the preset positive skewness threshold and the preset normal threshold. This is the first specific quantile value.

[0100] This is a weighting factor, with values ​​between 0 and 1, used to balance the contributions of the mean and quantiles. When the skewness coefficient is small, When the skewness coefficient is close to 1, the initial performance benchmark value is mainly determined by the average value. As the skewness coefficient increases, When the value is close to 0, the initial performance benchmark is mainly determined by the first specific quantile value.

[0101] Understandably, when the data distribution is close to normal, the skewness coefficient is close to 0, and the benchmark value depends more on the mean. When the data is positively skewed, the benchmark value depends more on the first specific quantile. The formula achieves a smooth transition through linear interpolation, ensuring that the benchmark value reflects both central tendency and distribution shape.

[0102] S25. If the standard deviation of the filtered data sequence is less than the first stability threshold, output the initial performance benchmark value as the final generated performance benchmark value.

[0103] S26. If the standard deviation is between the first stable threshold and the second stable threshold, the initial performance benchmark value is corrected based on the mean and the standard deviation, and the correction result is used as the final generated performance benchmark value.

[0104] It should be added that the first and second stability thresholds are dynamically adjusted based on the quantiles of the historical standard deviation distribution, such as the 25th and 75th quantiles.

[0105] It is understood that the correction formula corresponding to the aforementioned correction is: , This indicates the revised performance benchmark value. This is a conservative coefficient, which defaults to an empirical value, such as 0.1 or 0.15. The specific value is set by domain experts to control the degree of correction or conservatism.

[0106] in, It measures volatility; the greater the volatility, the larger this ratio. The loss rate is calculated by subtracting the loss rate from 1 and then combining it with the first specific quantile value, which effectively reduces the specific quantile value, resulting in a more conservative and robust performance benchmark value.

[0107] S27. When the standard deviation is greater than the second stability threshold, the second specific quantile value is used as the final performance benchmark value.

[0108] It's important to note that the calculation logic is adaptively switched based on the data standard deviation. Specifically, when the data is stable, the state-of-the-art value calculated based on leading samples is accepted. When the data exhibits moderate fluctuations, conservative adjustments are made to balance state-of-the-art performance with general applicability. When data shows significant differences, a more robust quantile is employed to avoid outlier interference. This ensures that the generated benchmark value is neither unattainable due to a few exceptionally high-performing units, nor loses its state-of-the-art value due to overall fluctuations or a few lagging units. It consistently provides a clear, reasonable, and achievable carbon emission performance improvement target for the majority of production units within the group.

[0109] In one specific embodiment, it should be noted that generating dynamic performance benchmark values ​​also includes setting upper and lower limits for the benchmark values ​​to ensure that the corrected benchmark value is not lower than a preset percentage of the historical best level of the peer group, and not higher than a preset percentage of the average level of the peer group. When the corrected calculated value exceeds the boundary, it is automatically truncated to the boundary value.

[0110] For example, the historical best level represents the limit that can be achieved under ideal conditions. The historical average level represents the general performance of the group. This ensures that the revised dynamic performance benchmark is no less than 90% of the historical best carbon intensity level of the peer group and no more than 110% of the historical average carbon intensity level of the peer group.

[0111] S3. Calculate the relative deviation between the real-time carbon emission intensity of a specified cement production unit and the corresponding performance benchmark value of its peer group.

[0112] S4. Obtain the historical relative deviation data of the designated cement production unit over multiple consecutive periods, and obtain the trend slope by analyzing its changing trend through linear regression.

[0113] Understandably, the least squares method is used to perform linear regression analysis on the historical relative deviation data series to calculate its trend slope. The trend slope represents the rate of change of relative deviation over time. In the least squares method, the time period number is set as the independent variable, the relative deviation is set as the dependent variable, and a linear equation is fitted.

[0114] S5. Combining the current relative deviation with the trend slope, a carbon emission exceedance trend index is calculated.

[0115] Please refer to the details. Figure 5 As shown, the specific calculation process of the carbon emission exceedance trend index includes steps S51-S54.

[0116] S51. Calculate the determination coefficient of the linear regression analysis and match the trend weight coefficient.

[0117] Understandably, when the coefficient of determination is less than 0.6, the trend weight coefficient is the coefficient of determination. When the coefficient of determination is greater than or equal to 0.6, the trend weight coefficient is assigned a value of 1. 0.6 is used as the current critical empirical value for the coefficient of determination, and this critical empirical value of 0.6 can be fine-tuned based on the volatility of the actual data, for example, taking a value within the range of 0.5-0.7.

[0118] S52. Multiply the trend weight coefficient by the trend slope to obtain the corrected trend slope, and determine whether it is greater than the preset slope.

[0119] Understandably, the preset slope represents the minimum critical rate at which the relative deviation deteriorates. When the corrected trend slope exceeds this value, it signifies that carbon emission performance is deteriorating at an unacceptable rate, requiring immediate attention and the initiation of compensation. Furthermore, the preset slope is determined based on the distribution of trend slopes in historical data of cases leading to carbon emission exceedances over a period prior to the exceedance, for example, by taking the 50th or 70th percentile of that distribution.

[0120] S53. If the slope is greater than the preset slope, the current relative deviation is compensated by matching the compensation coefficient based on the difference between the trend slope and the preset slope, and then the minimum-maximum normalization process is performed and the carbon emission exceedance trend index is output.

[0121] Understandably, compensating for the current relative deviation using the preset compensation coefficient can be divided into two cases: if the current relative deviation is greater than 0, then the compensation coefficient is multiplied by the current relative deviation. If the current relative deviation is less than 0, then the difference between the preset compensation coefficient and 1 is multiplied by the absolute value of the current relative deviation, and the result of the multiplication is added to the current relative deviation.

[0122] It should be noted that the compensation coefficient is positively correlated with the trend slope. For example, if the preset slope is 0.05 and the actual trend slope is 0.08, the compensation coefficient can be 1.2. If the difference increases to 0.06, the compensation coefficient can be increased to 1.5.

[0123] In the context of carbon emissions from cement production, the trend slope is typically small. When the preset slope is set to 0.05, the typical difference is... Concentrated in the range of 0-0.05. Through backtesting using historical data, a gradient compensation rule was set: when... When the value is in the range of 0 to 0.02, the compensation coefficient adopts a linear function. Slightly magnified, This is the compensation coefficient. When When it is in the range of 0.02 to 0.05, use Intensity amplification is performed. When When the deviation exceeds 0.05, the compensation coefficient is fixed at 1.5. This design achieves dynamic gradient compensation for the deviation of high-risk units by segmenting and amplifying the difference.

[0124] S54. If the slope is less than or equal to the preset slope, the current relative deviation will be processed by minimum-maximum normalization before being output.

[0125] It's important to note that a trend weighting coefficient filters out low-reliability random fluctuations, preventing false alarms. Secondly, when a deteriorating trend is both significant and exceeds a preset slope, a compensation mechanism amplifies the index value, prioritizing strong warnings for production units experiencing accelerated deterioration, even if their current absolute emission levels are not yet exceeding limits. This allows managers to precisely focus their attention on the most risky areas, shifting from reactive to proactive intervention, significantly improving the efficiency and effectiveness of carbon management.

[0126] S6. When the carbon emission exceedance trend index exceeds the preset risk threshold, an early warning signal is triggered and a management diagnostic report is generated.

[0127] Understandably, the preset risk threshold is determined through backtesting analysis of historical cases of exceeding the limit.

[0128] It should be noted that when the carbon emission exceedance trend index exceeds the preset risk threshold, a graded early warning signal will be automatically triggered, and a structured management diagnostic report will be generated simultaneously. This report accurately locates potential process links that lead to carbon emission exceedances by associating process tags in cluster analysis with real-time operational data, such as abnormal coal feeding in clinker calcination or energy efficiency degradation of grinding equipment. At the same time, it provides control priority suggestions based on the degree of deterioration of the trend slope, forming a closed-loop management support from early warning to decision-making.

[0129] The above content is merely an example and illustration of the concept of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the concept of the invention or exceed the scope defined by the present invention, and all such modifications and additions should fall within the protection scope of the present invention.

Claims

1. A method for evaluating carbon emissions of cement production based on digitalization driving, characterized in that, The method comprises: Real-time collection of operation characteristic data and carbon emission intensity of a plurality of cement production units, clustering analysis of the plurality of cement production units based on the operation characteristic data to divide the plurality of cement production units into a plurality of peer groups; Statistical analysis of historical carbon emission intensity data distribution of each cement production unit in each peer group, and dynamic generation of performance benchmark values of the peer group according to the statistical analysis, wherein the generation process comprises: For each peer group, collecting carbon emission intensity data sequences of all cement production units in the group within a preset statistical period; Calculating the standard deviation of the sequence, and if the standard deviation exceeds a preset stability threshold, screening carbon emission intensity data sequences of cement production units in the top percentage; Based on the screened data, calculating the average value, the standard deviation, the first specific quantile value and the second specific quantile value, the first specific quantile value being greater than the second specific quantile value, and determining the data distribution pattern; When the data is strongly positively skewed, the first specific quantile value is taken as the initial performance benchmark value; when the data is slightly positively skewed, the initial performance benchmark value is calculated based on a weighted combination of the first specific quantile value and the average value; when the data is close to normal, the average value is taken as the initial performance benchmark value; If the standard deviation of the screened data sequence is less than a first stability threshold, the initial performance benchmark value is output as the final generated performance benchmark value; If the standard deviation is between the first stability threshold and a second stability threshold, the initial performance benchmark value is modified based on the average value and the standard deviation, and the modified result is taken as the final generated performance benchmark value; When the standard deviation is greater than the second stability threshold, the second specific quantile value is taken as the final generated performance benchmark value; Calculating the relative deviation between the real-time carbon emission intensity of a specified cement production unit and the corresponding performance benchmark value of the peer group to which the specified cement production unit belongs; Obtaining historical relative deviation data of the specified cement production unit within a plurality of consecutive periods, and analyzing the trend of the historical relative deviation data by linear regression to obtain a trend slope; Combining the current relative deviation and the trend slope, the carbon emission over-standard trend index is calculated; When the carbon emission over-standard trend index exceeds a preset risk threshold, a warning signal is triggered and a management diagnosis report is generated; The preprocessed operation characteristic data is taken as a feature vector, and a differential weight coefficient is assigned to each feature dimension according to the production process to which the feature dimension belongs, and a weighted feature vector is constructed; a candidate cluster number set is preset according to the number range of typical process routes in the cement industry, for each candidate number, a clustering process is performed based on the weighted feature vector and a quality evaluation index is calculated; the candidate cluster number with the maximum quality evaluation index is taken as the target cluster number; a clustering process including process compatibility verification is performed based on the target cluster number and the weighted feature vector, and the mode of kiln type, mill type and capacity scale in each cluster is counted; a process label is generated in the format of capacity scale-kiln type-mill, and the cement production units are finally classified according to the process label.

2. The cement production carbon emission evaluation method based on digitalization driving according to claim 1, characterized in that: The operation characteristic data comprises production process parameters, material consumption data, yield data, equipment running time data and current production scheduling plan. The device runtime data includes device start-stop time points, maximum continuous running time, unplanned downtime frequency, unplanned downtime length, and production stage in which the unplanned downtime occurs.

3. The method for evaluating carbon emissions of cement production based on digitalization driving according to claim 2, characterized in that: The pre-processing of the operation feature data before clustering analysis further includes: The coal feeding amount fluctuation variance and kiln driving load fluctuation variance of the production unit are monitored, and quantile values of both in respective historical variance data sets are calculated; If the quantile values continuously fall below a first threshold value of the respective historical fluctuation variance for more than one complete production shift, it is determined that the current period is a stable working condition interval; The production process parameters, material consumption data, and yield data are Z-score standardized using the mean and standard deviation calculated from the historical data samples in the stable working condition interval, and standardized data is output; The device start-stop time points and the current production scheduling plan are input into a parser, time points are classified into corresponding production modes by matching event labels and timestamps, and a period mapping function is dynamically assigned to each mode; According to the production mode to which the device start-stop time point belongs, the corresponding period mapping function is called to calculate the transformed feature value; The unplanned downtime length and the production stage in which the unplanned downtime occurs are input, the downtime interference weight coefficient is calculated through downtime influence quantification, the unplanned downtime frequency and unplanned downtime length are linearly weighted based on the weight coefficient, and minimum-maximum normalization processing is performed; The standardized data, feature values, and normalization processing results are integrated to form a pre-processed data set.

4. The cement production carbon emission evaluation method based on digitalization driving according to claim 3, characterized in that: The allocation rules of the differentiated weight coefficients are: Assigning weights to characteristic dimensions belonging to the clinker calcination process ; Assigning weights to characteristic dimensions belonging to raw material preparation and cement grinding process ; Assigning weights to feature dimensions belonging within auxiliary procedures , and , .

5. The cement production carbon emission evaluation method based on digitalization driving according to claim 3, characterized in that: The specific calculation process of the quality evaluation index includes: For each candidate number in the candidate cluster number set, the number of initial cluster centers is selected based on cement production device configuration constraints, which include that the production units corresponding to the initial cluster centers are consistent in kiln type configuration and are in the same level in production capacity; The weighted Euclidean distance of each weighted feature vector to each cluster center is calculated, and iterative clustering is performed using the weighted Euclidean distance as the similarity measure; After each iteration, the production process compatibility in each cluster is detected, and the production units with process conflicts are re-assigned; If the clustering process converges, the average silhouette coefficient of the clustering result corresponding to the current cluster number and the variance of carbon emission intensity among clusters in the clustering result are calculated; Statistical significance test is performed on the variance of carbon emission intensity among clusters in the clustering result, and a test conclusion is obtained; If the statistical significance test fails, the quality evaluation index is assigned a value of zero; If the statistical significance test passes, the average silhouette coefficient and the variance are normalized, respectively, and the final quality evaluation index is output by linearly weighting and fusing.

6. The method for evaluating carbon emissions of cement production based on digitalization driving according to claim 1, characterized in that: The specific steps for determining the data distribution pattern include: The skewness coefficient is calculated based on the carbon emission intensity of the cement production unit, the mean and standard deviation of the filtered data sequence; When the skewness coefficient is greater than a preset positive skew threshold, it is determined to be a strong positive skew distribution; When the absolute value of the skewness coefficient is less than a preset normal threshold, it is determined to be close to a normal distribution; When the skewness coefficient is between the preset normal threshold and the preset positive skew threshold, it is determined to be a mild positive skew distribution.

7. The method for evaluating carbon emissions of cement production based on digitalization driving according to claim 5, characterized in that: The weighted mixing calculation formula is: , is an initial performance benchmark value, is a preset positive bias threshold value, is a skewness coefficient, represents a difference between the preset positive bias threshold value and a preset normal threshold value, is the average value, is a first specific quantile value.

8. The method for evaluating carbon emissions of cement production based on digitalization driving according to claim 1, characterized in that: The specific calculation process of the carbon emission over-standard tendency index comprises: calculating a determination coefficient of the linear regression analysis and matching a tendency weight coefficient; multiplying the tendency weight coefficient and the tendency slope to obtain a modified tendency slope, and judging whether the modified tendency slope is greater than a preset slope; if the modified tendency slope is greater than the preset slope, compensating a current relative deviation based on a difference between the tendency slope and the preset slope, matching a compensation coefficient, performing minimum-maximum normalization processing on the current relative deviation after the compensation, and outputting the carbon emission over-standard tendency index; if the modified tendency slope is less than or equal to the preset slope, performing minimum-maximum normalization processing on the current relative deviation and outputting.

Images